NeuralFields - Think in Fields - Build in Intelligence

Overview

Built-in Web

Smart routing & multi-LLM management API

LLM Routing normalizes every request and sends it to the model that fits the job, whether that is Claude, GPT, Gemini, Mistral, or a local Ollama deployment. Routing policies live in the gateway rather than your codebase, so you can move traffic to a cheaper model, pin a region for data residency, or roll back a bad routing change without shipping code. You keep your own provider contracts and keys, and every call is logged with its cost, latency, model, and the fallback path it took.

Route by cost, latency, quality, or region
Automatic failover across providers mid-request
Bring your own keys - no resold capacity

Before you pick this platform

Learn more about LLM Routing

Do I have to change my application code?

No. The router is API-compatible with the OpenAI and Anthropic SDKs, so in most cases you change the base URL and nothing else.

How much latency does routing add?

Under 10ms at p95. Routing decisions are made from cached health and pricing data, not a live lookup.

Can I self-host it?

Yes. It ships as a container you can run in your own VPC, with the managed control plane optional.

Overview

ElsaX Desktop

AI decision intelligence & interactive analytics

DataPulse records every LLM request with its cost, latency, model, and cache outcome, then lets you slice that by user, project, route, or any tag you attach. Drill from a chart down to the individual request behind a spike, compare what the same traffic would cost on a different model, and alert on budget overruns, p95 latency, or error rates through Slack, PagerDuty, or webhooks. Nothing is sampled, and everything exports to Grafana, Datadog, S3, or your own warehouse.

Slice cost and latency by user, project, or route
Alerts via Slack, PagerDuty, or webhooks
Export to Grafana, Datadog, or your data lake

Before you pick this platform

Learn more about DataPulse

Does DataPulse sample requests?

No. Every request is recorded with full cost and latency detail. Sampling hides exactly the outliers you need to see.

How does it get provider costs?

It reads token usage from each response and prices it against current provider rates, then reconciles against your actual invoices monthly.

Can I keep the data in my own infrastructure?

Yes. Self-hosted deployments write to your warehouse, and export streams to Grafana, Datadog, or S3 are available on every plan.

Overview

Built-in Web

Unified multi-cloud storage manager

Storage Gateway puts a single API and a single namespace in front of every bucket your team uses. Move a dataset between clouds without rewriting application paths, write retention and lifecycle rules once instead of once per console, and reclaim space with content-addressed dedup scoped safely per encryption boundary. Applications get short-lived, prefix-scoped credentials in place of long-lived cloud keys, and every read, write, and delete lands in an immutable audit log with the policy decision attached.

One namespace across S3, GCS, Azure Blob, and MinIO
Lifecycle, retention, and dedup policies applied once
Egress cut with edge caching and content addressing

Before you pick this platform

Learn more about Storage Gateway

Does my data pass through your servers?

No. The gateway issues scoped credentials and coordinates transfers, but object data moves directly between your applications and your buckets.

What happens if the gateway is unavailable?

Your buckets keep working. Clients can fall back to direct cloud access, and cached credentials remain valid for their lifetime.

Is on-prem storage supported?

Yes. Any S3-compatible target works, including MinIO, Ceph, and NetApp StorageGRID.

Overview

ElsaX Desktop

Multi-tenant, spec-driven API gateway & MCP integration suite

API Explorer reads your OpenAPI definition and turns it into two things at once: a governed gateway with routing, auth, validation, and rate limits, and a matching MCP tool set your agents can call directly. Tool names and descriptions are derived from the spec, so they cannot drift as the API evolves, and HTTP, gRPC, WebSocket, and SSE endpoints are all served from the same definition. Per-agent scopes decide which operation each caller may reach, enforced at the gateway rather than trusted to a prompt, and every invocation is recorded with its arguments, result, and latency.

Generate MCP tools straight from your OpenAPI spec
HTTP, gRPC, WebSocket, MCP, and SSE from one definition
Per-agent scopes, request validation, and full call logs

Before you pick this platform

Learn more about API Explorer

Which spec versions are supported?

OpenAPI 3.0 and 3.1. Swagger 2.0 specs are converted on import.

Do I still need my existing API gateway?

Not usually. It handles auth, validation, and rate limiting, but it can also sit behind an existing gateway if you prefer.

How are breaking spec changes handled?

Tools are versioned. A breaking change publishes a new tool version while the old one keeps working until you retire it.

Overview

Formant Desktop

Autonomous robotics & drone swarm orchestration

Formant is the control plane for autonomous fleets. Define a mission once and it spreads the work across available drones and ground robots, handling hand-off when a unit returns to charge and keeping one shared map every unit reads from. SLAM, obstacle avoidance, and detection run on-board with no GPS or cloud dependency, so a dropped link pauses your dashboard rather than the mission. Operators supervise from an exception-first console, take manual control of any unit mid-mission, and replay a whole sortie afterwards with the full sensor record.

Multi-unit mission planning with automatic hand-off
On-device perception that survives network drops
Live telemetry, teleop, and full mission replay

Before you pick this platform

Learn more about Formant

Which hardware is supported?

Any unit running PX4, ArduPilot, or ROS 2 can pair through the edge agent, including most Jetson- and Qualcomm-based platforms.

Does it need constant connectivity?

No. Missions run fully on-device. Telemetry buffers locally and syncs when the link returns.

Can operators take manual control?

Yes. Any unit can be switched to teleop mid-mission and handed back to autonomy without restarting the mission.

Overview

Web Plugins

Collaborative task & workflow automation platform

FlowDesk keeps tasks, reviews, and approvals in the same workspace as the automations that move them along, so a handoff never waits on someone remembering to ping. Any step can be assigned to a person or to an agent, and both pass through the same required reviewers before the work advances. Every automated action records the inputs it ran on, the output it produced, and the policy that allowed it, so an audit is a filter rather than an excavation. Cycle-time reporting shows where work waits, which tells you what to automate next.

Assign any step to a person or an agent
Automations that move work without manual pings
Every automated action logged with its inputs

Before you pick this platform

Learn more about FlowDesk

Is this another ticket tracker?

It models processes rather than issues. Teams commonly keep their tracker for engineering work and run cross-functional pipelines in FlowDesk.

How much control do agents get?

Exactly what you grant per step. Agent actions can require the same approvals a person would, and can be limited to specific tools and data.

Can I import existing boards?

Yes. CSV and Jira imports bring over tasks, assignees, and status history.

Overview

Web Core

Centralized project management hub for AI & software lifecycles

Workspace is the control center for a project: the people working on it, the code repositories behind it, the AI skills and tools it can reach, and the API endpoints it deploys, all held together in one place. Instead of tracking members in one system, code in another, and deployments in a third, a project carries its whole surface at once — so adding an engineer, granting a tool, or shipping an endpoint happens against the same record rather than three disconnected ones. Access follows the project, so a person or an agent added once gets the repositories, tools, and endpoints that project allows, and loses them the moment they leave.

Members, repos, tools, and endpoints on one project record
Access granted once and revoked everywhere at once
Humans and agents governed by the same project roles

Before you pick this platform

Learn more about Workspace

Does this replace our issue tracker?

No. It holds the project's people, code, tools, and deployments - the surface a project has. Day-to-day task tracking stays wherever your team already does it.

How do agents fit in?

An agent joins a project the same way a person does, with a role that decides which repositories, tools, and endpoints it may reach.

What happens when someone leaves a project?

Removing them from the project revokes their access to every surface that project covers, in one action, with the change recorded.

Overview

Docker Container

An edge AI runtime that unifies software engineering, game development, data science, and autonomous systems

ElsaX works where your team already does: the repo, the build, the pipeline. Describe a change and it opens the files it needs, proposes a complete staged diff with its reasoning next to each hunk, and runs your test command before asking for approval. Nothing is written until you accept it. The same review gates cover asset pipelines and NPC scripting for game studios and transformations and backfills for data teams, while multi-model routing sends each task to whichever model is cheapest and strongest for that kind of work. Prompt, model, diff, and outcome are recorded for every action it takes.

Staged diffs with one-click apply or reject
Runs your tests before asking for approval
Multi-model routing with a full audit trail

Before you pick this platform

Learn more about ElsaX Runtime

Can it modify files without asking?

Only if you opt in with an explicit auto-apply flag. By default every change is staged and waits for your approval.

Does it work in CI?

Yes. It runs headless with structured output, so you can use it for automated refactors, migration checks, or review comments.

How does model routing decide?

Each task type has a policy that weighs cost, latency, and past success rate. You can pin a specific model per project or per command.

Every product in the NeuralFields stack

Explore the platforms

Smart routing & multi-LLM management API

AI decision intelligence & interactive analytics

Unified multi-cloud storage manager

Multi-tenant, spec-driven API gateway & MCP integration suite

Autonomous robotics & drone swarm orchestration

Collaborative task & workflow automation platform

Centralized project management hub for AI & software lifecycles

An edge AI runtime that unifies software engineering, game development, data science, and autonomous systems

All 8 platforms

Not sure which platform you need?